Your daily AI usage
What that looks like
Water figures based on Li et al. (2023), "Making AI Less 'Thirsty'", UC Riverside. Text query: ~25 ml/exchange; image generation: ~100 ml/image; code completion: ~8 ml/suggestion; voice: ~12 ml/minute. Li et al. count the water used to cool the servers plus the water used to generate their electricity, for GPT-3 in Microsoft's U.S. data centers: a 500 ml bottle for roughly 10 to 50 responses. Individual figures vary by data center and model.
For a whole facility, the data center water calculator estimates a data center's daily water use from its megawatts. For the electricity behind the same prompts, see how much energy ChatGPT uses per prompt.
Why does AI use so much water?
AI models run on GPU clusters that generate enormous heat. Data centers cool those servers using chilled water systems — and that water evaporates into the atmosphere, it doesn't return to local watersheds. A single large data center can consume 1–5 million gallons of water per day.
The problem is location: most hyperscale AI data centers are built in places like Arizona, Nevada, and Texas — states already under severe water stress — because land is cheap and power is available. Local water utilities in these regions have reported that large data center contracts can squeeze out allocation for residential growth, agriculture, and municipal reserves. The communities bearing the water cost rarely see economic benefits proportional to that burden — data centers are capital-intensive, largely automated facilities that employ few local workers relative to their resource draw.
Training a single large AI model can consume as much water as 700,000 liters — equivalent to manufacturing 370 cars. And that's before a single user query is answered.
Common questions about AI and water use
Water use during AI inference (running queries) is separate from the much larger water cost of training. GPT-3's training alone consumed an estimated 700,000 liters. The calculator's 25 ml text-query rate comes from Li et al., who count both cooling water at the data center and the water used to generate its electricity.